A Neural-Net based PID Controllers for Nonlinear Multivariable Systems

M. Tokuda, T. Yamamoto, and Y. Monden


Neural network, nonlinear system, PID control, generalized predictive control, multivariable systems, decoupling


In process industries, PID control schemes have been popular because of their simple structures and ease of comprehension of the physical meanings of the control gains. However, good control performance cannot be obtained by simply using PID control schemes, as most processes are considered as nonlinear multivariable systems with mutual interactions. The authors propose a design method of neural-net based PID controllers for nonlinear multivariable systems with mutual interactions. The proposed method consists of a static decoupler that plays a role of roughly decoupling the controlled object, multiloop PID controllers, and some multilayered neural networks used in order to improve decoupling and linearize the approximately decoupled controlled object. Finally, the effectiveness of the proposed control scheme is evaluated on some simulation examples.

Important Links:

Go Back